Meta releases Muse Wonder and charges $4 per million input tokens for it on its API

Meta released Muse Wonder on October 12, 2026, and opened it the same day on the Meta Model API, its pay-as-you-go service for developers, under the model ID muse-wonder-1.0. The company describes it as its most capable model, built for complex coding work and for agents, programs that carry out a task over many steps on their own. A stronger variant, Muse Wonder Full, is available only inside Muse Code, Meta's coding tool, and only with a paid subscription.

What it costs on the API

On the API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens. Input served from the prompt cache costs $0.20 per million tokens. Meta adds no surcharge for long prompts, so you pay the same rate whether the context window is nearly empty or nearly full.

Muse Spark 1.3, Meta's smaller model, costs $1.25 per million input tokens and $4.25 per million output tokens. By our calculation, that makes Muse Wonder 3.2 times as expensive per input token and about 4.7 times as expensive per output token. Both models take the same requests on the same endpoints, so moving a job from one to the other is a change of model ID.

One discount does not carry over. Muse Spark has a Contributor mode with much lower prices, in exchange for letting Meta train future models on your prompts and responses. On the API, Muse Wonder is offered only in Standard mode, where Meta's documentation says your content is not used for training.

What the model accepts

Muse Wonder accepts text, images, video, audio, and PDF files, and it answers in text. Its context window, the amount of material the model can consider in one request, is 1,048,576 tokens. Meta labels the model a preview, and its documentation says availability varies by country.

Muse Wonder Full stays inside Muse Code

You cannot call Muse Wonder Full through the API, and setting Muse Wonder to its highest reasoning effort does not route a request to it. The Full variant is included only in the Power Usage and Maximum Usage plans of Muse Code. Meta's documentation does not list prices for those plans.

Muse Code subscriptions start in Contributor mode by default. In that mode Meta may use the prompts, code, files, and responses from your sessions to train future models, and you get more usage in return. You can switch it off in settings at any time.

What Meta claims, and what it has not shown yet

Meta says the two models reach state-of-the-art results on benchmarks for personal and professional agents, memory, and coding. Its model card compares them with the strongest models from OpenAI and Anthropic at maximum reasoning effort, and notes that its runs of competitor models were best-effort and may not have been optimally tuned. No independent measurement has been published yet.

The same model card says that, without safeguards, Meta does not rule out Muse Wonder Full reaching the high-risk threshold set in its own safety framework in the chemical and biological domain. With safety training and classifiers in place, the company rates the risk as moderate or lower in every domain it tested. A full safety report is still to come, and the model card states that the model is not intended for people under 18.

Meta also says Muse Wonder will soon power the Muse app, reaching subscribers first, but it has not given a date.